Monthly Total Number of alarms

The total number of alarms of this month is :

• 36991 alarms were detected.

This represents a decrease of -32459 number of alarms from the previous month’s total of 69450 number of alarms.

Monthly sub-categories of alarms

Out of this total number of alarms, we may further categorize alarms into :

• 36816 alarms, are classified as “Chattering/Fleeting”. This means that these alarms are coming from a chattering meter (a meter which frequently/constantly hits the system by sending alarms). Some of these may last a certain duration, and they may also recur on a regular basis (e.g., a regular 10 minute interval pattern). Others may have no self-evident pattern (e.g., may appear in a fleeting way, turning “on” and “off” very quickly, with no discernable regularity in terms of timespan between alarm occurences).

• 36991 alarms, are classified as “Flood periods”. This means that more than 10 alarms appeared, and several meters were sending alarms during this 10 minute period.

• 81 alarms, are classified as “Stale alarms”. This means that these alarms lasted longer than 24h.

A representative value is given by the ISA18.2 “Test <1% of 10 minute periods contain more than 10 alarms”

The percentage of 10 minutes periods which have more than 10 alarms of this month is :

• 100 % detected.

This represents no change from the previous month with a stable percentage of 100 from the previous month’s percentage of 100 10 minutes periods containing more than 10 alarms.

Here is a graph showing a daily nr of

Monthly Alarms Rates per 10 minutes interval

This provides a more detailed view

The key element here is the duration of the stale alarms

These stale alarms often are main contributors to the “constant baseline” of alarms.

What is important to understand is that the duration of the sale alarms is the key metric.

• If for example some “over-THD alarm” are triggered by 15 different meters, and if the over THD issue lasted for 20 days :

– The alarms system already has 20 days of constant presence of alarm floods.

– This is considered a “constant baseline” during these 20 days.

This type of events is visually very easy to detect on a graph, as this functions like a “step” when turned “on” and then later when it is turened “off”.

Here is a graph illustrating the impact of the stale alarms on this month’s presence of alarms in each 10 minute timestamp.

This represents a reduction of 65 % of the presence of alarms in all the 10 minutes timestamps.

How would the removal or aggregation of STALE alarms improve the number of present alamrs

So if we were able to remove the stale alarms, here is what the monthly number of alarms would look like for each 10 minutes timestamp.

These alarms are due to the constant re-occuring conditions and can be named “constant-chattering alarms”

These “constant-chattering alarms” often also main contributors to the “constant baseline” of alarms.

In this case, what is important to understand is that the 2 key combined metrics are :

• The PRESENCE : This is the number of 10 minutes timestamps where the alarm is present

• The FREQUENCY : This is the number of alarms, so the number of occurences per each of these 10 minutes timestamps.

If for example 5 meters pick up a “THD voltage alarm” condition which appears for a short duration on an average of 5 times in every one of the 10 minutes intervals during that month, the system again would end up with a constant average number of alarms of 25 alarms per 10 minutes time interval.

Here is a graph illustrating the impact of the constant alarms on this month’s presence of alarms in each 10 minute timestamp.

This represents a reduction of 65 % of the presence of alarms in all the 10 minutes timestamps. CHANGE VARIABLE TO CONSTANT CHATTERING

How would the removal or aggregation of CONSTANT CHATTERING alarms improve the number of present alamrs

So if we were able to remove the stale alarms, here is what the monthly number of alarms would look like for each 10 minutes timestamp.

These alarms are due to the constant re-occuring conditions and can be named “constant-chattering alarms”

These “constant-chattering alarms” often also main contributors to the “constant baseline” of alarms.

In this case, what is important to understand is that the 2 key combined metrics are :

• The PRESENCE : This is the number of 10 minutes timestamps where the alarm is present

• The FREQUENCY : This is the number of alarms, so the number of occurences per each of these 10 minutes timestamps.

If for example 5 meters pick up a “THD voltage alarm” condition which appears for a short duration on an average of 5 times in every one of the 10 minutes intervals during that month, the system again would end up with a constant average number of alarms of 25 alarms per 10 minutes time interval.

Here is a graph illustrating the impact of the constant alarms on this month’s presence of alarms in each 10 minute timestamp.

This represents a reduction of 65 % of the presence of alarms in all the 10 minutes timestamps.

How would the removal or aggregation of FLEETING alarms improve the number of present alamrs

So if we were able to remove the stale alarms, here is what the monthly number of alarms would look like for each 10 minutes timestamp.

Monthly spread of sources of all alarms

An important metric to look at on a monthly basis is the number of Sources generating new/recurring alarms.

• List of Top 10 sources generating the alarms

– With their number of alarms generated per source

– And the respective percentage of all alarms generated per source

• This provides an idea of the spread of new/recurring issues :

– Maybe only very few devices generate 80% of all new/recurring alarms. In this case, it’s easy to focus the attention on these sources.

– Or, maybe all devices capable of picking up a specific type of alarm capture it. In this case, the event may be generated upstream of all these devices (e.g., a Voltage sag, coming from the utility, or a welding manufacturer next to the monitored facility ?).

This then needs to be compare with the total number of sources which generated at least one alarm during this month :

• Below’s table shows the total number of sources of all alarms, as well as a splitt per type of generated alarms (number of chattering, flood, stale)

And the next table shows the same, but by relative percentages :

Monthly spread of sources of the “Stale” alarms

Another linked metric to look at is the source of lasting alarms (alarms which last longer than 10 minutes).

• Some alarms will stay “on” for a several 10 minutes intervals (e.G., several alarms on harmonics/THD levels above acceptable threshold, measured by 5 different meters in a specific location, lasting for 8 hours in an industrial site, during a specific process step).

– This increases the constant number of alarms during such extended periods, and may increase the number of 10 minute periods with more than 10 alarms.

– In terms of recommendations, this may help focus investigation on these meters and time periods leveraging the types of alarms and the correlation of these meters and alarms as a group.

The following table will show the list of SourceIds generating more than 0.5 % of all new/recurring issues (percentage of the number of different new alarms of this month)

Top 10 or 50 List ranked by decreasing presence & duration, duration of each, priority grouping, etc.

This table will help understand what are the most frequent and present stale alarms.

Filtering out or aggregating the alarms which regurlarly or constantly hit your system

• Your system has 31303 identified as being “regular / constant alarms”

• These “regular / constant alarms” represent :

– As much as 50.3 percent of the 10 minutes timestamps.

– Yet they only represent 5.3 percent of the total number of alarms.

– This means… that each one of these alarms is much more present than most of the other alarms (by a ratio of 9.4 times more present than the other alarms).

We recommend to start identifying these types of chattering alarms. The reason is simple : These alarms will be present in many 10 minutes intervals, and so will be main contributors to the baseline (the minimum number of alarms occurring per 10 minutes interval)

And to do this we recommend using a threshold of 8.3 percent of all the 10 minutes interval to determine this type of alarms.

This threshold will detect alarms which hit your system at least once every 2 hours. This 2 hours threshold is important as some alarms may be calculated over a given time interval (such as some average values). And typically, these intervals last 10 minutes or 1 hour, but in certain cases, up to 2 hours. So, by using this 2 hour threshold, this table will accurately capture these frequently occurring alarms.

So you could reduce the number of alarms by

• Filtering these alarms (they would become 0 alarms). Before you do this, you must analyze these constantly hitting alarms to understand if these are false alarms (e.G., such as when a threshold is set to wrong threshold or wrong nominal voltage). You should not remove any alarms without doing any analysis.

• Or by aggregating these alarms and adding a counter number counting the number of alarms and/or a number of 10 minutes intervals where these alarms are present during this month (each alarm would become 1 alarm). The advantage of creating such a table is that you keep all the information visible and available for analysis in a much more synthetic way, and you avoid unnecessarily cluttering the other alarms with these cosntanty occuring alarms.

Here is an example of such an aggregation table :

Here is how the total number of alarms would look like if you applied this recommended aggregation :

Number of fleeting & low priority alarms,

List ranked by decreasing presence & duration, duration of each, priority grouping, etc.

Number of chattering & HIGH priority alarms,

List ranked by decreasing presence & duration, duration of each, priority grouping, etc.

This shows a simplified table of all alarms for each day

This should be changed to a daily metric (average Nr of alarm per day)

Number of alarms per 10 minute period

In_Progress (description text)

Number of alarms per hourly period

In_Progress (description text)

Summary table

• as numbers

– Top 10 (sum(freq))

– More than 10 (%)

– chattering (%)

– Flood (%)

– Stale (future) (%)

List table (PRIO 1) => 1 table

Summary table

• Most frequent ALARM - List (Top 10)

– Top 10 (sum(freq))

– More than 10 (%)

– chattering (%)

– Flood (%)

– Stale (future) (%)

List table (PRIO 1) => 1 table

• Top 50 Alarms (with SourceIds)

– List by decreasing rank

– as alarms text

— Cause_text, Effect_text, SourceId, Freq, Rank (group_by SourceId, Alarm_Text, ….Freq(on Nr of ID)) … NOT cause_value NOT effect_value